blob: daaf866316478dd262f344124f02c9891530dc7b [file] [log] [blame]
Rusty Russell2e04ef72009-07-30 16:03:45 -06001/*P:800
2 * Interrupts (traps) are complicated enough to earn their own file.
Rusty Russellf938d2c2007-07-26 10:41:02 -07003 * There are three classes of interrupts:
4 *
5 * 1) Real hardware interrupts which occur while we're running the Guest,
6 * 2) Interrupts for virtual devices attached to the Guest, and
7 * 3) Traps and faults from the Guest.
8 *
9 * Real hardware interrupts must be delivered to the Host, not the Guest.
10 * Virtual interrupts must be delivered to the Guest, but we make them look
11 * just like real hardware would deliver them. Traps from the Guest can be set
12 * up to go directly back into the Guest, but sometimes the Host wants to see
13 * them first, so we also have a way of "reflecting" them into the Guest as if
Rusty Russell2e04ef72009-07-30 16:03:45 -060014 * they had been delivered to it directly.
15:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -070016#include <linux/uaccess.h>
Rusty Russellc18acd72007-10-22 11:03:35 +100017#include <linux/interrupt.h>
18#include <linux/module.h>
Alexey Dobriyand43c36d2009-10-07 17:09:06 +040019#include <linux/sched.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070020#include "lg.h"
21
Rusty Russellc18acd72007-10-22 11:03:35 +100022/* Allow Guests to use a non-128 (ie. non-Linux) syscall trap. */
23static unsigned int syscall_vector = SYSCALL_VECTOR;
24module_param(syscall_vector, uint, 0444);
25
Rusty Russellbff672e2007-07-26 10:41:04 -070026/* The address of the interrupt handler is split into two bits: */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070027static unsigned long idt_address(u32 lo, u32 hi)
28{
29 return (lo & 0x0000FFFF) | (hi & 0xFFFF0000);
30}
31
Rusty Russell2e04ef72009-07-30 16:03:45 -060032/*
33 * The "type" of the interrupt handler is a 4 bit field: we only support a
34 * couple of types.
35 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070036static int idt_type(u32 lo, u32 hi)
37{
38 return (hi >> 8) & 0xF;
39}
40
Rusty Russellbff672e2007-07-26 10:41:04 -070041/* An IDT entry can't be used unless the "present" bit is set. */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -030042static bool idt_present(u32 lo, u32 hi)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070043{
44 return (hi & 0x8000);
45}
46
Rusty Russell2e04ef72009-07-30 16:03:45 -060047/*
48 * We need a helper to "push" a value onto the Guest's stack, since that's a
49 * big part of what delivering an interrupt does.
50 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020051static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070052{
Rusty Russellbff672e2007-07-26 10:41:04 -070053 /* Stack grows upwards: move stack then write value. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070054 *gstack -= 4;
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020055 lgwrite(cpu, *gstack, u32, val);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070056}
57
Rusty Russell2e04ef72009-07-30 16:03:45 -060058/*H:210
59 * The set_guest_interrupt() routine actually delivers the interrupt or
Rusty Russellbff672e2007-07-26 10:41:04 -070060 * trap. The mechanics of delivering traps and interrupts to the Guest are the
61 * same, except some traps have an "error code" which gets pushed onto the
62 * stack as well: the caller tells us if this is one.
63 *
64 * "lo" and "hi" are the two parts of the Interrupt Descriptor Table for this
65 * interrupt or trap. It's split into two parts for traditional reasons: gcc
66 * on i386 used to be frightened by 64 bit numbers.
67 *
68 * We set up the stack just like the CPU does for a real interrupt, so it's
69 * identical for the Guest (and the standard "iret" instruction will undo
Rusty Russell2e04ef72009-07-30 16:03:45 -060070 * it).
71 */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -030072static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
73 bool has_err)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070074{
Rusty Russell47436aa2007-10-22 11:03:36 +100075 unsigned long gstack, origstack;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070076 u32 eflags, ss, irq_enable;
Rusty Russell47436aa2007-10-22 11:03:36 +100077 unsigned long virtstack;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070078
Rusty Russell2e04ef72009-07-30 16:03:45 -060079 /*
80 * There are two cases for interrupts: one where the Guest is already
Rusty Russellbff672e2007-07-26 10:41:04 -070081 * in the kernel, and a more complex one where the Guest is in
Rusty Russell2e04ef72009-07-30 16:03:45 -060082 * userspace. We check the privilege level to find out.
83 */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -020084 if ((cpu->regs->ss&0x3) != GUEST_PL) {
Rusty Russell2e04ef72009-07-30 16:03:45 -060085 /*
86 * The Guest told us their kernel stack with the SET_STACK
87 * hypercall: both the virtual address and the segment.
88 */
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -020089 virtstack = cpu->esp1;
90 ss = cpu->ss1;
Rusty Russell47436aa2007-10-22 11:03:36 +100091
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -020092 origstack = gstack = guest_pa(cpu, virtstack);
Rusty Russell2e04ef72009-07-30 16:03:45 -060093 /*
94 * We push the old stack segment and pointer onto the new
Rusty Russellbff672e2007-07-26 10:41:04 -070095 * stack: when the Guest does an "iret" back from the interrupt
96 * handler the CPU will notice they're dropping privilege
Rusty Russell2e04ef72009-07-30 16:03:45 -060097 * levels and expect these here.
98 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020099 push_guest_stack(cpu, &gstack, cpu->regs->ss);
100 push_guest_stack(cpu, &gstack, cpu->regs->esp);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700101 } else {
Rusty Russellbff672e2007-07-26 10:41:04 -0700102 /* We're staying on the same Guest (kernel) stack. */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200103 virtstack = cpu->regs->esp;
104 ss = cpu->regs->ss;
Rusty Russell47436aa2007-10-22 11:03:36 +1000105
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200106 origstack = gstack = guest_pa(cpu, virtstack);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700107 }
108
Rusty Russell2e04ef72009-07-30 16:03:45 -0600109 /*
110 * Remember that we never let the Guest actually disable interrupts, so
Rusty Russellbff672e2007-07-26 10:41:04 -0700111 * the "Interrupt Flag" bit is always set. We copy that bit from the
Rusty Russelle1e72962007-10-25 15:02:50 +1000112 * Guest's "irq_enabled" field into the eflags word: we saw the Guest
Rusty Russell2e04ef72009-07-30 16:03:45 -0600113 * copy it back in "lguest_iret".
114 */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200115 eflags = cpu->regs->eflags;
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200116 if (get_user(irq_enable, &cpu->lg->lguest_data->irq_enabled) == 0
Rusty Russelle5faff42007-07-20 22:11:13 +1000117 && !(irq_enable & X86_EFLAGS_IF))
118 eflags &= ~X86_EFLAGS_IF;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700119
Rusty Russell2e04ef72009-07-30 16:03:45 -0600120 /*
121 * An interrupt is expected to push three things on the stack: the old
Rusty Russellbff672e2007-07-26 10:41:04 -0700122 * "eflags" word, the old code segment, and the old instruction
Rusty Russell2e04ef72009-07-30 16:03:45 -0600123 * pointer.
124 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200125 push_guest_stack(cpu, &gstack, eflags);
126 push_guest_stack(cpu, &gstack, cpu->regs->cs);
127 push_guest_stack(cpu, &gstack, cpu->regs->eip);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700128
Rusty Russellbff672e2007-07-26 10:41:04 -0700129 /* For the six traps which supply an error code, we push that, too. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700130 if (has_err)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200131 push_guest_stack(cpu, &gstack, cpu->regs->errcode);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700132
Rusty Russell2e04ef72009-07-30 16:03:45 -0600133 /*
134 * Now we've pushed all the old state, we change the stack, the code
135 * segment and the address to execute.
136 */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200137 cpu->regs->ss = ss;
138 cpu->regs->esp = virtstack + (gstack - origstack);
139 cpu->regs->cs = (__KERNEL_CS|GUEST_PL);
140 cpu->regs->eip = idt_address(lo, hi);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700141
Rusty Russell2e04ef72009-07-30 16:03:45 -0600142 /*
143 * There are two kinds of interrupt handlers: 0xE is an "interrupt
144 * gate" which expects interrupts to be disabled on entry.
145 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700146 if (idt_type(lo, hi) == 0xE)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200147 if (put_user(0, &cpu->lg->lguest_data->irq_enabled))
148 kill_guest(cpu, "Disabling interrupts");
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700149}
150
Rusty Russelle1e72962007-10-25 15:02:50 +1000151/*H:205
Rusty Russellbff672e2007-07-26 10:41:04 -0700152 * Virtual Interrupts.
153 *
Rusty Russellabd41f02009-06-12 22:27:02 -0600154 * interrupt_pending() returns the first pending interrupt which isn't blocked
155 * by the Guest. It is called before every entry to the Guest, and just before
Rusty Russell2e04ef72009-07-30 16:03:45 -0600156 * we go to sleep when the Guest has halted itself.
157 */
Rusty Russella32a8812009-06-12 22:27:02 -0600158unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700159{
160 unsigned int irq;
161 DECLARE_BITMAP(blk, LGUEST_IRQS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700162
Rusty Russellbff672e2007-07-26 10:41:04 -0700163 /* If the Guest hasn't even initialized yet, we can do nothing. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200164 if (!cpu->lg->lguest_data)
Rusty Russellabd41f02009-06-12 22:27:02 -0600165 return LGUEST_IRQS;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700166
Rusty Russell2e04ef72009-07-30 16:03:45 -0600167 /*
168 * Take our "irqs_pending" array and remove any interrupts the Guest
169 * wants blocked: the result ends up in "blk".
170 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200171 if (copy_from_user(&blk, cpu->lg->lguest_data->blocked_interrupts,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700172 sizeof(blk)))
Rusty Russellabd41f02009-06-12 22:27:02 -0600173 return LGUEST_IRQS;
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200174 bitmap_andnot(blk, cpu->irqs_pending, blk, LGUEST_IRQS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700175
Rusty Russellbff672e2007-07-26 10:41:04 -0700176 /* Find the first interrupt. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700177 irq = find_first_bit(blk, LGUEST_IRQS);
Rusty Russella32a8812009-06-12 22:27:02 -0600178 *more = find_next_bit(blk, LGUEST_IRQS, irq+1);
Rusty Russellabd41f02009-06-12 22:27:02 -0600179
180 return irq;
181}
182
Rusty Russell2e04ef72009-07-30 16:03:45 -0600183/*
184 * This actually diverts the Guest to running an interrupt handler, once an
185 * interrupt has been identified by interrupt_pending().
186 */
Rusty Russella32a8812009-06-12 22:27:02 -0600187void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more)
Rusty Russellabd41f02009-06-12 22:27:02 -0600188{
189 struct desc_struct *idt;
190
191 BUG_ON(irq >= LGUEST_IRQS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700192
Rusty Russell2e04ef72009-07-30 16:03:45 -0600193 /*
194 * They may be in the middle of an iret, where they asked us never to
195 * deliver interrupts.
196 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200197 if (cpu->regs->eip >= cpu->lg->noirq_start &&
198 (cpu->regs->eip < cpu->lg->noirq_end))
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700199 return;
200
Rusty Russellbff672e2007-07-26 10:41:04 -0700201 /* If they're halted, interrupts restart them. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200202 if (cpu->halted) {
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700203 /* Re-enable interrupts. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200204 if (put_user(X86_EFLAGS_IF, &cpu->lg->lguest_data->irq_enabled))
205 kill_guest(cpu, "Re-enabling interrupts");
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200206 cpu->halted = 0;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700207 } else {
Rusty Russellbff672e2007-07-26 10:41:04 -0700208 /* Otherwise we check if they have interrupts disabled. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700209 u32 irq_enabled;
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200210 if (get_user(irq_enabled, &cpu->lg->lguest_data->irq_enabled))
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700211 irq_enabled = 0;
Rusty Russella32a8812009-06-12 22:27:02 -0600212 if (!irq_enabled) {
213 /* Make sure they know an IRQ is pending. */
214 put_user(X86_EFLAGS_IF,
215 &cpu->lg->lguest_data->irq_pending);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700216 return;
Rusty Russella32a8812009-06-12 22:27:02 -0600217 }
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700218 }
219
Rusty Russell2e04ef72009-07-30 16:03:45 -0600220 /*
221 * Look at the IDT entry the Guest gave us for this interrupt. The
Rusty Russellbff672e2007-07-26 10:41:04 -0700222 * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip
Rusty Russell2e04ef72009-07-30 16:03:45 -0600223 * over them.
224 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200225 idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq];
Rusty Russellbff672e2007-07-26 10:41:04 -0700226 /* If they don't have a handler (yet?), we just ignore it */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700227 if (idt_present(idt->a, idt->b)) {
Rusty Russellbff672e2007-07-26 10:41:04 -0700228 /* OK, mark it no longer pending and deliver it. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200229 clear_bit(irq, cpu->irqs_pending);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600230 /*
231 * set_guest_interrupt() takes the interrupt descriptor and a
Rusty Russellbff672e2007-07-26 10:41:04 -0700232 * flag to say whether this interrupt pushes an error code onto
Rusty Russell2e04ef72009-07-30 16:03:45 -0600233 * the stack as well: virtual interrupts never do.
234 */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300235 set_guest_interrupt(cpu, idt->a, idt->b, false);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700236 }
Rusty Russell6c8dca52007-07-27 13:42:52 +1000237
Rusty Russell2e04ef72009-07-30 16:03:45 -0600238 /*
239 * Every time we deliver an interrupt, we update the timestamp in the
Rusty Russell6c8dca52007-07-27 13:42:52 +1000240 * Guest's lguest_data struct. It would be better for the Guest if we
241 * did this more often, but it can actually be quite slow: doing it
242 * here is a compromise which means at least it gets updated every
Rusty Russell2e04ef72009-07-30 16:03:45 -0600243 * timer interrupt.
244 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200245 write_timestamp(cpu);
Rusty Russella32a8812009-06-12 22:27:02 -0600246
Rusty Russell2e04ef72009-07-30 16:03:45 -0600247 /*
248 * If there are no other interrupts we want to deliver, clear
249 * the pending flag.
250 */
Rusty Russella32a8812009-06-12 22:27:02 -0600251 if (!more)
252 put_user(0, &cpu->lg->lguest_data->irq_pending);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700253}
Rusty Russell9f155a92009-06-12 22:27:08 -0600254
255/* And this is the routine when we want to set an interrupt for the Guest. */
256void set_interrupt(struct lg_cpu *cpu, unsigned int irq)
257{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600258 /*
259 * Next time the Guest runs, the core code will see if it can deliver
260 * this interrupt.
261 */
Rusty Russell9f155a92009-06-12 22:27:08 -0600262 set_bit(irq, cpu->irqs_pending);
263
Rusty Russell2e04ef72009-07-30 16:03:45 -0600264 /*
265 * Make sure it sees it; it might be asleep (eg. halted), or running
266 * the Guest right now, in which case kick_process() will knock it out.
267 */
Rusty Russell9f155a92009-06-12 22:27:08 -0600268 if (!wake_up_process(cpu->tsk))
269 kick_process(cpu->tsk);
270}
Rusty Russellc18acd72007-10-22 11:03:35 +1000271/*:*/
272
Rusty Russell2e04ef72009-07-30 16:03:45 -0600273/*
274 * Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent
Rusty Russellc18acd72007-10-22 11:03:35 +1000275 * me a patch, so we support that too. It'd be a big step for lguest if half
276 * the Plan 9 user base were to start using it.
277 *
278 * Actually now I think of it, it's possible that Ron *is* half the Plan 9
Rusty Russell2e04ef72009-07-30 16:03:45 -0600279 * userbase. Oh well.
280 */
Rusty Russellc18acd72007-10-22 11:03:35 +1000281static bool could_be_syscall(unsigned int num)
282{
283 /* Normal Linux SYSCALL_VECTOR or reserved vector? */
284 return num == SYSCALL_VECTOR || num == syscall_vector;
285}
286
287/* The syscall vector it wants must be unused by Host. */
288bool check_syscall_vector(struct lguest *lg)
289{
290 u32 vector;
291
292 if (get_user(vector, &lg->lguest_data->syscall_vec))
293 return false;
294
295 return could_be_syscall(vector);
296}
297
298int init_interrupts(void)
299{
300 /* If they want some strange system call vector, reserve it now */
Yinghai Lub77b8812008-12-19 15:23:44 -0800301 if (syscall_vector != SYSCALL_VECTOR) {
302 if (test_bit(syscall_vector, used_vectors) ||
303 vector_used_by_percpu_irq(syscall_vector)) {
304 printk(KERN_ERR "lg: couldn't reserve syscall %u\n",
305 syscall_vector);
306 return -EBUSY;
307 }
308 set_bit(syscall_vector, used_vectors);
Rusty Russellc18acd72007-10-22 11:03:35 +1000309 }
Yinghai Lub77b8812008-12-19 15:23:44 -0800310
Rusty Russellc18acd72007-10-22 11:03:35 +1000311 return 0;
312}
313
314void free_interrupts(void)
315{
316 if (syscall_vector != SYSCALL_VECTOR)
317 clear_bit(syscall_vector, used_vectors);
318}
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700319
Rusty Russell2e04ef72009-07-30 16:03:45 -0600320/*H:220
321 * Now we've got the routines to deliver interrupts, delivering traps like
Rusty Russella6bd8e12008-03-28 11:05:53 -0500322 * page fault is easy. The only trick is that Intel decided that some traps
Rusty Russell2e04ef72009-07-30 16:03:45 -0600323 * should have error codes:
324 */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300325static bool has_err(unsigned int trap)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700326{
327 return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17);
328}
329
Rusty Russellbff672e2007-07-26 10:41:04 -0700330/* deliver_trap() returns true if it could deliver the trap. */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300331bool deliver_trap(struct lg_cpu *cpu, unsigned int num)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700332{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600333 /*
334 * Trap numbers are always 8 bit, but we set an impossible trap number
335 * for traps inside the Switcher, so check that here.
336 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200337 if (num >= ARRAY_SIZE(cpu->arch.idt))
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300338 return false;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700339
Rusty Russell2e04ef72009-07-30 16:03:45 -0600340 /*
341 * Early on the Guest hasn't set the IDT entries (or maybe it put a
342 * bogus one in): if we fail here, the Guest will be killed.
343 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200344 if (!idt_present(cpu->arch.idt[num].a, cpu->arch.idt[num].b))
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300345 return false;
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200346 set_guest_interrupt(cpu, cpu->arch.idt[num].a,
347 cpu->arch.idt[num].b, has_err(num));
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300348 return true;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700349}
350
Rusty Russell2e04ef72009-07-30 16:03:45 -0600351/*H:250
352 * Here's the hard part: returning to the Host every time a trap happens
Rusty Russellbff672e2007-07-26 10:41:04 -0700353 * and then calling deliver_trap() and re-entering the Guest is slow.
Rusty Russelle1e72962007-10-25 15:02:50 +1000354 * Particularly because Guest userspace system calls are traps (usually trap
355 * 128).
Rusty Russellbff672e2007-07-26 10:41:04 -0700356 *
357 * So we'd like to set up the IDT to tell the CPU to deliver traps directly
358 * into the Guest. This is possible, but the complexities cause the size of
359 * this file to double! However, 150 lines of code is worth writing for taking
360 * system calls down from 1750ns to 270ns. Plus, if lguest didn't do it, all
Rusty Russelle1e72962007-10-25 15:02:50 +1000361 * the other hypervisors would beat it up at lunchtime.
Rusty Russellbff672e2007-07-26 10:41:04 -0700362 *
Rusty Russell56adbe92007-10-22 11:03:28 +1000363 * This routine indicates if a particular trap number could be delivered
Rusty Russell2e04ef72009-07-30 16:03:45 -0600364 * directly.
365 */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300366static bool direct_trap(unsigned int num)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700367{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600368 /*
369 * Hardware interrupts don't go to the Guest at all (except system
370 * call).
371 */
Rusty Russellc18acd72007-10-22 11:03:35 +1000372 if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num))
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300373 return false;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700374
Rusty Russell2e04ef72009-07-30 16:03:45 -0600375 /*
376 * The Host needs to see page faults (for shadow paging and to save the
Rusty Russellbff672e2007-07-26 10:41:04 -0700377 * fault address), general protection faults (in/out emulation) and
Matias Zabaljauregui4cd8b5e2009-03-14 13:37:52 -0200378 * device not available (TS handling), invalid opcode fault (kvm hcall),
Rusty Russell2e04ef72009-07-30 16:03:45 -0600379 * and of course, the hypercall trap.
380 */
Matias Zabaljauregui4cd8b5e2009-03-14 13:37:52 -0200381 return num != 14 && num != 13 && num != 7 &&
382 num != 6 && num != LGUEST_TRAP_ENTRY;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700383}
Rusty Russellf56a3842007-07-26 10:41:05 -0700384/*:*/
385
Rusty Russell2e04ef72009-07-30 16:03:45 -0600386/*M:005
387 * The Guest has the ability to turn its interrupt gates into trap gates,
Rusty Russellf56a3842007-07-26 10:41:05 -0700388 * if it is careful. The Host will let trap gates can go directly to the
389 * Guest, but the Guest needs the interrupts atomically disabled for an
390 * interrupt gate. It can do this by pointing the trap gate at instructions
Rusty Russell2e04ef72009-07-30 16:03:45 -0600391 * within noirq_start and noirq_end, where it can safely disable interrupts.
392 */
Rusty Russellf56a3842007-07-26 10:41:05 -0700393
Rusty Russell2e04ef72009-07-30 16:03:45 -0600394/*M:006
395 * The Guests do not use the sysenter (fast system call) instruction,
Rusty Russellf56a3842007-07-26 10:41:05 -0700396 * because it's hardcoded to enter privilege level 0 and so can't go direct.
397 * It's about twice as fast as the older "int 0x80" system call, so it might
398 * still be worthwhile to handle it in the Switcher and lcall down to the
399 * Guest. The sysenter semantics are hairy tho: search for that keyword in
Rusty Russell2e04ef72009-07-30 16:03:45 -0600400 * entry.S
401:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700402
Rusty Russell2e04ef72009-07-30 16:03:45 -0600403/*H:260
404 * When we make traps go directly into the Guest, we need to make sure
Rusty Russellbff672e2007-07-26 10:41:04 -0700405 * the kernel stack is valid (ie. mapped in the page tables). Otherwise, the
406 * CPU trying to deliver the trap will fault while trying to push the interrupt
407 * words on the stack: this is called a double fault, and it forces us to kill
408 * the Guest.
409 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600410 * Which is deeply unfair, because (literally!) it wasn't the Guests' fault.
411 */
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200412void pin_stack_pages(struct lg_cpu *cpu)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700413{
414 unsigned int i;
415
Rusty Russell2e04ef72009-07-30 16:03:45 -0600416 /*
417 * Depending on the CONFIG_4KSTACKS option, the Guest can have one or
418 * two pages of stack space.
419 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200420 for (i = 0; i < cpu->lg->stack_pages; i++)
Rusty Russell2e04ef72009-07-30 16:03:45 -0600421 /*
422 * The stack grows *upwards*, so the address we're given is the
Rusty Russell8057d762007-08-30 06:35:08 +1000423 * start of the page after the kernel stack. Subtract one to
424 * get back onto the first stack page, and keep subtracting to
Rusty Russell2e04ef72009-07-30 16:03:45 -0600425 * get to the rest of the stack pages.
426 */
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200427 pin_page(cpu, cpu->esp1 - 1 - i * PAGE_SIZE);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700428}
429
Rusty Russell2e04ef72009-07-30 16:03:45 -0600430/*
431 * Direct traps also mean that we need to know whenever the Guest wants to use
Rusty Russellbff672e2007-07-26 10:41:04 -0700432 * a different kernel stack, so we can change the IDT entries to use that
433 * stack. The IDT entries expect a virtual address, so unlike most addresses
434 * the Guest gives us, the "esp" (stack pointer) value here is virtual, not
435 * physical.
436 *
437 * In Linux each process has its own kernel stack, so this happens a lot: we
Rusty Russell2e04ef72009-07-30 16:03:45 -0600438 * change stacks on each context switch.
439 */
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200440void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700441{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600442 /*
443 * You're not allowed a stack segment with privilege level 0: bad Guest!
444 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700445 if ((seg & 0x3) != GUEST_PL)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200446 kill_guest(cpu, "bad stack segment %i", seg);
Rusty Russellbff672e2007-07-26 10:41:04 -0700447 /* We only expect one or two stack pages. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700448 if (pages > 2)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200449 kill_guest(cpu, "bad stack pages %u", pages);
Rusty Russellbff672e2007-07-26 10:41:04 -0700450 /* Save where the stack is, and how many pages */
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200451 cpu->ss1 = seg;
452 cpu->esp1 = esp;
453 cpu->lg->stack_pages = pages;
Rusty Russellbff672e2007-07-26 10:41:04 -0700454 /* Make sure the new stack pages are mapped */
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200455 pin_stack_pages(cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700456}
457
Rusty Russell2e04ef72009-07-30 16:03:45 -0600458/*
459 * All this reference to mapping stacks leads us neatly into the other complex
460 * part of the Host: page table handling.
461 */
Rusty Russellbff672e2007-07-26 10:41:04 -0700462
Rusty Russell2e04ef72009-07-30 16:03:45 -0600463/*H:235
464 * This is the routine which actually checks the Guest's IDT entry and
465 * transfers it into the entry in "struct lguest":
466 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200467static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700468 unsigned int num, u32 lo, u32 hi)
469{
470 u8 type = idt_type(lo, hi);
471
Rusty Russellbff672e2007-07-26 10:41:04 -0700472 /* We zero-out a not-present entry */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700473 if (!idt_present(lo, hi)) {
474 trap->a = trap->b = 0;
475 return;
476 }
477
Rusty Russellbff672e2007-07-26 10:41:04 -0700478 /* We only support interrupt and trap gates. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700479 if (type != 0xE && type != 0xF)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200480 kill_guest(cpu, "bad IDT type %i", type);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700481
Rusty Russell2e04ef72009-07-30 16:03:45 -0600482 /*
483 * We only copy the handler address, present bit, privilege level and
Rusty Russellbff672e2007-07-26 10:41:04 -0700484 * type. The privilege level controls where the trap can be triggered
485 * manually with an "int" instruction. This is usually GUEST_PL,
Rusty Russell2e04ef72009-07-30 16:03:45 -0600486 * except for system calls which userspace can use.
487 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700488 trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF);
489 trap->b = (hi&0xFFFFEF00);
490}
491
Rusty Russell2e04ef72009-07-30 16:03:45 -0600492/*H:230
493 * While we're here, dealing with delivering traps and interrupts to the
Rusty Russellbff672e2007-07-26 10:41:04 -0700494 * Guest, we might as well complete the picture: how the Guest tells us where
495 * it wants them to go. This would be simple, except making traps fast
496 * requires some tricks.
497 *
498 * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the
Rusty Russell2e04ef72009-07-30 16:03:45 -0600499 * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here.
500 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200501void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700502{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600503 /*
504 * Guest never handles: NMI, doublefault, spurious interrupt or
505 * hypercall. We ignore when it tries to set them.
506 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700507 if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY)
508 return;
509
Rusty Russell2e04ef72009-07-30 16:03:45 -0600510 /*
511 * Mark the IDT as changed: next time the Guest runs we'll know we have
512 * to copy this again.
513 */
Glauber de Oliveira Costaae3749d2008-01-17 19:14:46 -0200514 cpu->changed |= CHANGED_IDT;
Rusty Russellbff672e2007-07-26 10:41:04 -0700515
Rusty Russell56adbe92007-10-22 11:03:28 +1000516 /* Check that the Guest doesn't try to step outside the bounds. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200517 if (num >= ARRAY_SIZE(cpu->arch.idt))
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200518 kill_guest(cpu, "Setting idt entry %u", num);
Rusty Russell56adbe92007-10-22 11:03:28 +1000519 else
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200520 set_trap(cpu, &cpu->arch.idt[num], num, lo, hi);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700521}
522
Rusty Russell2e04ef72009-07-30 16:03:45 -0600523/*
524 * The default entry for each interrupt points into the Switcher routines which
Rusty Russellbff672e2007-07-26 10:41:04 -0700525 * simply return to the Host. The run_guest() loop will then call
Rusty Russell2e04ef72009-07-30 16:03:45 -0600526 * deliver_trap() to bounce it back into the Guest.
527 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700528static void default_idt_entry(struct desc_struct *idt,
529 int trap,
Rusty Russell0c120912008-07-29 09:58:31 -0500530 const unsigned long handler,
531 const struct desc_struct *base)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700532{
Rusty Russellbff672e2007-07-26 10:41:04 -0700533 /* A present interrupt gate. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700534 u32 flags = 0x8e00;
535
Rusty Russell2e04ef72009-07-30 16:03:45 -0600536 /*
537 * Set the privilege level on the entry for the hypercall: this allows
538 * the Guest to use the "int" instruction to trigger it.
539 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700540 if (trap == LGUEST_TRAP_ENTRY)
541 flags |= (GUEST_PL << 13);
Rusty Russell0c120912008-07-29 09:58:31 -0500542 else if (base)
Rusty Russell2e04ef72009-07-30 16:03:45 -0600543 /*
544 * Copy privilege level from what Guest asked for. This allows
545 * debug (int 3) traps from Guest userspace, for example.
546 */
Rusty Russell0c120912008-07-29 09:58:31 -0500547 flags |= (base->b & 0x6000);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700548
Rusty Russellbff672e2007-07-26 10:41:04 -0700549 /* Now pack it into the IDT entry in its weird format. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700550 idt->a = (LGUEST_CS<<16) | (handler&0x0000FFFF);
551 idt->b = (handler&0xFFFF0000) | flags;
552}
553
Rusty Russellbff672e2007-07-26 10:41:04 -0700554/* When the Guest first starts, we put default entries into the IDT. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700555void setup_default_idt_entries(struct lguest_ro_state *state,
556 const unsigned long *def)
557{
558 unsigned int i;
559
560 for (i = 0; i < ARRAY_SIZE(state->guest_idt); i++)
Rusty Russell0c120912008-07-29 09:58:31 -0500561 default_idt_entry(&state->guest_idt[i], i, def[i], NULL);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700562}
563
Rusty Russell2e04ef72009-07-30 16:03:45 -0600564/*H:240
565 * We don't use the IDT entries in the "struct lguest" directly, instead
Rusty Russellbff672e2007-07-26 10:41:04 -0700566 * we copy them into the IDT which we've set up for Guests on this CPU, just
Rusty Russell2e04ef72009-07-30 16:03:45 -0600567 * before we run the Guest. This routine does that copy.
568 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200569void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700570 const unsigned long *def)
571{
572 unsigned int i;
573
Rusty Russell2e04ef72009-07-30 16:03:45 -0600574 /*
575 * We can simply copy the direct traps, otherwise we use the default
576 * ones in the Switcher: they will return to the Host.
577 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200578 for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) {
Rusty Russell0c120912008-07-29 09:58:31 -0500579 const struct desc_struct *gidt = &cpu->arch.idt[i];
580
Rusty Russell56adbe92007-10-22 11:03:28 +1000581 /* If no Guest can ever override this trap, leave it alone. */
582 if (!direct_trap(i))
583 continue;
584
Rusty Russell2e04ef72009-07-30 16:03:45 -0600585 /*
586 * Only trap gates (type 15) can go direct to the Guest.
Rusty Russell56adbe92007-10-22 11:03:28 +1000587 * Interrupt gates (type 14) disable interrupts as they are
588 * entered, which we never let the Guest do. Not present
Rusty Russell0c120912008-07-29 09:58:31 -0500589 * entries (type 0x0) also can't go direct, of course.
590 *
591 * If it can't go direct, we still need to copy the priv. level:
592 * they might want to give userspace access to a software
Rusty Russell2e04ef72009-07-30 16:03:45 -0600593 * interrupt.
594 */
Rusty Russell0c120912008-07-29 09:58:31 -0500595 if (idt_type(gidt->a, gidt->b) == 0xF)
596 idt[i] = *gidt;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700597 else
Rusty Russell0c120912008-07-29 09:58:31 -0500598 default_idt_entry(&idt[i], i, def[i], gidt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700599 }
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700600}
601
Rusty Russelle1e72962007-10-25 15:02:50 +1000602/*H:200
603 * The Guest Clock.
604 *
605 * There are two sources of virtual interrupts. We saw one in lguest_user.c:
606 * the Launcher sending interrupts for virtual devices. The other is the Guest
607 * timer interrupt.
608 *
609 * The Guest uses the LHCALL_SET_CLOCKEVENT hypercall to tell us how long to
610 * the next timer interrupt (in nanoseconds). We use the high-resolution timer
611 * infrastructure to set a callback at that time.
612 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600613 * 0 means "turn off the clock".
614 */
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200615void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700616{
617 ktime_t expires;
618
619 if (unlikely(delta == 0)) {
620 /* Clock event device is shutting down. */
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200621 hrtimer_cancel(&cpu->hrt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700622 return;
623 }
624
Rusty Russell2e04ef72009-07-30 16:03:45 -0600625 /*
626 * We use wallclock time here, so the Guest might not be running for
Rusty Russelle1e72962007-10-25 15:02:50 +1000627 * all the time between now and the timer interrupt it asked for. This
Rusty Russell2e04ef72009-07-30 16:03:45 -0600628 * is almost always the right thing to do.
629 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700630 expires = ktime_add_ns(ktime_get_real(), delta);
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200631 hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700632}
633
Rusty Russelle1e72962007-10-25 15:02:50 +1000634/* This is the function called when the Guest's timer expires. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700635static enum hrtimer_restart clockdev_fn(struct hrtimer *timer)
636{
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200637 struct lg_cpu *cpu = container_of(timer, struct lg_cpu, hrt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700638
Rusty Russelle1e72962007-10-25 15:02:50 +1000639 /* Remember the first interrupt is the timer interrupt. */
Rusty Russell9f155a92009-06-12 22:27:08 -0600640 set_interrupt(cpu, 0);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700641 return HRTIMER_NORESTART;
642}
643
Rusty Russelle1e72962007-10-25 15:02:50 +1000644/* This sets up the timer for this Guest. */
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200645void init_clockdev(struct lg_cpu *cpu)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700646{
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200647 hrtimer_init(&cpu->hrt, CLOCK_REALTIME, HRTIMER_MODE_ABS);
648 cpu->hrt.function = clockdev_fn;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700649}